Journal of Biosocial Science http://journals.cambridge.org/JBS Additional services for Journal
of Biosocial Science:
Email alerts: Click here Subscriptions: Click here Commercial reprints: Click here Terms of use : Click here
REGIONAL VARIATIONS IN INFANT AND CHILD MORTALITY IN NIGERIA: A MULTILEVEL ANALYSIS SUNDAY A. ADEDINI, CLIFFORD ODIMEGWU, EUNICE N. S. IMASIKU, DOROTHY N. ONONOKPONO and LATIFAT IBISOMI Journal of Biosocial Science / Volume 47 / Issue 02 / March 2015, pp 165 - 187 DOI: 10.1017/S0021932013000734, Published online: 10 January 2014
Link to this article: http://journals.cambridge.org/abstract_S0021932013000734 How to cite this article: SUNDAY A. ADEDINI, CLIFFORD ODIMEGWU, EUNICE N. S. IMASIKU, DOROTHY N. ONONOKPONO and LATIFAT IBISOMI (2015). REGIONAL VARIATIONS IN INFANT AND CHILD MORTALITY IN NIGERIA: A MULTILEVEL ANALYSIS. Journal of Biosocial Science, 47, pp 165-187 doi:10.1017/S0021932013000734 Request Permissions : Click here
Downloaded from http://journals.cambridge.org/JBS, IP address: 131.187.94.93 on 18 Mar 2015
J. Biosoc. Sci., (2015) 47, 165–187, 6 Cambridge University Press, 2014 doi:10.1017/S0021932013000734 First published online 10 Jan 2014
REGIONAL VARIATIONS IN INFANT AND C H I L D M O R T A L I T Y I N NI G E R I A : A M U L T IL E V E L A N A L Y S I S SUNDAY A. ADEDINI*†1, CLIFFORD ODIMEGWU*, EUNICE N. S. IMASIKU*‡, DOROTHY N. ONONOKPONO*§ and LATIFAT IBISOMI* *Demography and Population Studies Programme, Schools of Public Health and Social Sciences, University of the Witwatersrand, Johannesburg, South Africa, †Demography and Social Statistics Department, Obafemi Awolowo University, Ile-Ife, Nigeria, ‡Department of Geography, University of Zambia, Lusaka, Zambia and §Department of Sociology and Anthropology, University of Uyo, Nigeria Summary. There are substantial regional disparities in under-five mortality in Nigeria, and evidence suggests that both individual- and community-level characteristics have an influence on health outcomes. Using 2008 Nigeria Demographic and Health Survey data, this study (1) examines the effects of individual- and community-level characteristics on infant/child mortality in Nigeria and (2) determines the extent to which characteristics at these levels influence regional variations in infant/child mortality in the country. Multilevel Cox proportional hazard analysis was performed on a nationally representative sample of 28,647 children nested within 18,028 mothers of reproductive age, who were also nested within 886 communities. The results indicate that community-level variables (such as region, place of residence, community infrastructure, community hospital delivery and community poverty level) and individual-level factors (including child’s sex, birth order, birth interval, maternal education, maternal age and wealth index) are important determinants of infant/child mortality in Nigeria. For instance, the results show a lower risk of death in infancy for children of mothers residing in communities with a high proportion of hospital delivery (HR: 0.70, p < 0.05) and for children whose mothers had secondary or higher education (HR: 0.84, p < 0.05). Although community factors appear to influence the association between individual-level factors and death during infancy and childhood, the findings consistently indicate that community-level characteristics are more important in explaining regional variations in child mortality, while individual-level factors are more important for regional variations in infant mortality. The results of this study underscore the need to look beyond the influence of individual-level factors in addressing regional variations in infant and child mortality in Nigeria. 1
Corresponding author. Email: [email protected]
165
166
S. A. Adedini et al. Introduction
Despite modest improvements in child health outcomes during the 20th century, infant and child mortality rates remain unacceptably high in the sub-Saharan African countries. Globally, around 7 million under-five deaths were recorded in 2011 (UNICEF, 2012). Sub-Saharan Africa is a major contributor to this statistic as more than two in five under-five deaths occur in the region (Black et al., 2003; Rutherford et al., 2010). Scholars have attempted to examine the factors influencing this. For instance, studies have established a significant relationship between infant and child mortality and individual-level characteristics such as maternal education, wealth status and other socioeconomic characteristics (Adetunji, 1995; Zaba & David, 1996; Lawoyin, 2001; Buor, 2002; Odimegwu, 2002; Fayeun & Omololu, 2011). Other studies have shown that attributes of the community context tend to influence the health outcomes of individuals (Sastry, 1996; Uthman, 2008; Babalola & Fatusi, 2009; Adekanmbi et al., 2013). This suggests that living in an economically and socially deprived community or neighbourhood could bring about an increase in mortality risk. Nigeria – the most populous country in Africa – is characterized by socially and economically advantaged and disadvantaged regions. The country is geographically, religiously, socially, ecologically and economically diverse. This has led to varied disease exposures and different health outcomes (Wall, 1998; Lawoyin, 2001; Grais et al., 2007). From the arid northern region of the country to the savannah west, from the predominantly Islamic north-west to the south-east vastly dominated by Christians, the country is highly heterogeneous and diverse. As a result, there is a huge diversity in the regional environment, cultural practices (Antai et al., 2009), health-seeking practices (Babalola & Fatusi, 2009), socioeconomic status (Aremu et al., 2011) and the political milieu. To date, Nigeria has a very high rate of under-five mortality, as about one in every six children born in the country dies before the age of five (NPC & ICF Macro, 2009). Worse still, infant and child mortality rates vary substantially from one region of the country to the other. For instance, under-five mortality rate ranges from 89 per 1000 live births in the south-west to 222 per 1000 live births in the north-east. Besides, Nigeria is not making sufficient progress towards the attainment of Millennium Development Goal four (MDG-4), as the under-five mortality rate in the country for the 1998–2003 period was 201 per 1000 live births, while the rate marginally declined to 157 per 1000 live births during the 2004–2008 period (NPC & ORC Macro, 2004; NPC & ICF Macro, 2009). In addition, as noted earlier, Nigeria is by far the most populous country in Africa and has a very huge childhood population. According to the 2006 population and housing census, Nigerian’s population is 140,431,790 and the population of the under-five children is 16.1% of the total population (NPC & ICF Macro, 2009). Furthermore, despite the improvement in medical technology, reports by NPC & ICF Macro (2009) indicate that Nigeria is still faced with several health challenges. These include high rates of under-five mortality (157 per 1000 live births), high teenage pregnancy (23% of young women aged 15–19 have already given birth), many poor pregnancy outcomes (such as stillbirth, spontaneous abortion and low birth weight), poor survival chances for the newborn and high unmet need for family planning (20% of currently married women have an unmet need for contraception in the country).
Infant and child mortality in Nigeria
167
Also, the percentage of births assisted by unskilled birth attendants is 60% in the country. The list of the country’s public health problems is endless. While many studies have been conducted on childhood mortality in Nigeria, as in other developing countries, Nigerian studies have paid more attention to the influence of individual-level attributes, and less attention to community-level determinants of childhood mortality. However, the literature shows that knowledge about the determinants of childhood mortality at the individual level is insufficient to address the problem (Sastry, 1996; Whitworth & Stephenson, 2002; Griffiths et al., 2004; Harttgen & Misselhorn, 2006; Antai, 2011b). This is because the contextual characteristics of the community or neighbourhood where children are born or raised tend to modify childlevel and mother-level characteristics and therefore affect children survival chances. In addition, past Nigerian studies that have examined community determinants of child survival have focused mainly on ages 0–59 months, thereby ignoring what the effects of community characteristics on child survival are during infancy (i.e. age 0–11 months) and childhood (12–59 months). To this end, this study aims to (1) examine the effects of individual- and community-level characteristics on child survival during infancy and childhood, and (2) determine the extent to which characteristics at the individual and community levels explain regional variations in infant and child mortality in Nigeria. Theoretical background From a theoretical standpoint, Mosley and Chen’s 1984 model on the proximate causes of childhood mortality establishes a relationship between child survival and determinants at various levels of operation: individual, household and community levels (WHO, 2003). Diez-Roux et al. (2001) posited that the physical and social characteristics of the neighbourhood where a person lives may affect health and health-related behaviour. Galster’s (2010) work on the mechanisms of neighbourhood effects theory observed a link between residential environment and the health outcomes of individual adults and children residing in such environment or community. The effect of community context, where children are born or raised, on their survival chances has been widely recognized (Sastry, 1997b; Omariba et al., 2007; Antai, 2009). Evidence suggests that living in an economically and socially deprived community or neighbourhood is associated with an increased risk of under-five mortality (Aremu et al., 2011). For instance, children born or raised in a community that lacks electricity, good drinking water and health facilities are likely to suffer from the same deprivation that can directly or indirectly influence their health outcomes. Further, Manda (2001) opined that Demographic and Health Surveys (DHSs) often collect birth history data that are clustered at the household and community levels. Sastry (1997b) also maintained that most demographic surveys conducted in developing countries often collect survival data that are clustered at both family and community levels. Omariba and colleagues (2007) argued that since DHSs collect child survival data from individual mothers in sampled households, then the children of those mothers cannot be regarded as independent observations. This is as a result of natural clustering or as a result of the data collection procedures (i.e. two-stage cluster sampling design) used in DHS data collection (Sastry, 1997b). The children from sampled households who are also nested within the same individual mother’s level tend to share similar characteristics and common genetic
168
S. A. Adedini et al.
factors. This is also true of children from the same community. Individuals in the same community are likely to be more homogenous than those from different communities. Similarities are expected in the health outcomes of children who are exposed to the same environmental conditions. By contrast, differences are expected in health outcomes of children raised in different communities due to differences in community characteristics (Harttgen & Misselhorn, 2006). A distinction is drawn between children living in a relatively better-off neighbourhood and those living in a relatively worse-off neighbourhood (Macintyre et al., 2002). Children living in two different households with similar socioeconomic characteristics can suffer different mortality risks if they are from two contrasting communities. Sastry (1996) contended that community characteristics can exacerbate or mitigate mortality risks of individuals depending on the environmental context the individuals find themselves. Griffiths (2004) also argued that community services and levels of infrastructural development of a community are capable of amplifying or reducing mortality risks among the children. This is because an individual child resident in a household unit, which in turn is located within a community, is exposed to various levels (within the societal hierarchy) that either directly or indirectly influence his or her survival chances. In addition, Whitworth & Stephenson (2002) maintained that two neonates with similar characteristics may suffer different neonatal mortality risks because of the community contextual effects. The authors argued that these differentials in mortality risks may be as a result of differences in the provision of antenatal and obstetric health care or the effects of environmental conditions the children are exposed to. Also, individuals residing in the same community tend to share similar preferences, cultural practices, values and customs. The reason is that individuals with similar tastes and values tend to cluster and live together (Sastry, 1997a). All this clustering and living together of people with common norms, values, identities and cultural practices, as well as spatial inequality in infrastructural development (Antai, 2011b), has a direct or indirect effect on the health outcomes of under-five children and this often brings about differentials in health outcomes between communities, particularly communities with contrasting characteristics. Motivated by the emerging interest in the study of effects of community or neighbourhood context on health outcomes in developing countries, this paper seeks to examine the extent to which characteristics at individual and community levels influence regional variations in infant and child mortality in Nigeria. Because previous attempts at examining regional disparities in child survival in Nigeria have focused mainly on the period within age 0–59 months, the main focus of the present analysis was the regional disparities in child survival during the two distinct periods of infancy (0–11 months) and childhood (12–59 months). Data and Methods Data source This study utilizes the birth recode of the 2008 Nigeria Demographic and Health Survey (NDHS) data. The survey elicited information on demographic and health indicators. The primary sampling unit (PSU), which was regarded as a cluster for the
Infant and child mortality in Nigeria
169
2008 NDHS, was defined on the basis of enumeration areas (EAs). The sample for the survey was selected using a stratified two-stage cluster design consisting of 888 clusters (NPC & ICF Macro, 2009). In all, a nationally representative sample of 36,800 households was selected for the survey. The community-level variables were measured at the level of the PSU, which serves as a proxy for community or neighbourhood. The PSUs refer to small and administratively defined areas and one PSU is a cluster consisting of at least 80 fairly homogenous households or units. In the 2008 NDHS, data were collected in 886 PSUs, while two PSUs could not be accessed due to disturbances in the areas. A full report of the data collection procedures for the 2008 NDHS is available elsewhere (NPC & ICF Macro, 2009). In the 2008 NDHS, birth history data were collected from 33,385 women aged 15–49 years. These included sex of child, month and year of child’s birth, child’s survivorship status, child’s current age and age at death if the child had died. Analysis for the present paper was restricted to 28,647 live-born children produced by 18,028 women within the five years before the survey. Analysis was restricted to births within the five years preceding the survey to obtain a picture of the current situations in the various geo-political regions of the country. However, severe omission of births and deaths as well as displacement of dates of those events could seriously affect mortality estimates. Meanwhile, the data quality assessment of the 2008 NDHS indicates that the percentage of missing information regarding births, deaths, birth dates as well as age at death across various characteristics of mothers such as mother’s education and region of residence only varied between less than 1% and 3% (NPC & ICF Macro, 2009). This suggests that there was neither a serious under-reporting of infant and child deaths nor serious displacement of information on such vital events that could seriously affect the mortality estimates across regions or across any other mother-level characteristic. In addition, to ensure national representativeness, weighting factors were applied to adjust for oversampling of some locations and under-sampling of others in the Demographic and Health Survey. Ethical considerations This study was based on secondary analysis of an existing dataset with all participant identifiers removed. The survey instruments received ethical approval from the National Ethics Committee in the Federal Ministry of Health, Abuja, Nigeria, and from the Ethics Committee of the Opinion Research Corporation of Macro International Inc., Calverton, MD, USA. Permission to use the 2008 Nigeria DHS data for this study was obtained from ICF Macro Inc. Outcome variables The outcome variables for this study were the risks of death in infancy or childhood, measured as the duration of survival since birth in months. These are defined as either the risk of a child dying between birth and first birthday (infant mortality) or between age 12 and 59 months (child mortality). Analysis was child-based and restricted to the live births in the 5 years before the survey. Hence, all children born within the 5 years before the survey date were included in the analysis. The children’s survival status and the age
170
S. A. Adedini et al.
at death in months (if the child had died) or the last month they were known to be alive (if the child was still living at the time of the survey) were combined to generate the outcome variables for the survival analysis. Taking the two durations (0–11 months and 12–59 months) into consideration, children known to have died (i.e. non-censored) were regarded as the cases, whereas children who were still alive at the time of the survey were treated as right-censored (as appropriate for infant and child mortality). Exposure variables Community contextual factors. As established in the reviewed literature, the community contextual characteristics of interest in this study included: (1) place of residence, categorized as (i) urban, (ii) rural; (2) region of residence, defined as the regions where the children were raised, and categorized as (i) South-West, (ii) North-Central, (iii) North-East, (iv) North-West, (v) South-East, and (vi) South-South; (3) community poverty level, defined as the average poverty level in the community, and categorized as (i) low, (ii) middle and (iii) high. Other contextual characteristics were (4) community maternal level of education, defined as the proportion of mothers who had at least secondary level of education in the community and categorized as (i) low, (ii) middle, (iii) high; (5) community hospital delivery, defined as proportion of mothers who had hospital delivery in the community, and categorized as (i) low, (ii) middle (iii) high; (6) community prenatal care by skilled provider, defined as the proportion of mothers who attended prenatal care by a skilled health provider in the community and categorized as (i) low, (ii) middle (iii) high; (7) proportion with electric connection in the community, defined as the proportion of mothers from households with electric connection in the community and categorized as (i) low, (ii) middle (iii) high; and (8) proportion with piped water in the community, defined as the proportion of mothers from households that had piped water as source of drinking water in the community and categorized as (i) low, (ii) middle (iii) high. Individual-level characteristics. Based on the reviewed literature, the important individual-level (i.e. child-level and mother-level) characteristics considered in this study were as follows: (1) birth order, defined as birth order of the child and categorized as (i) first births, (ii) 2–4 birth order, (iii) birth order 5þ; (2) child’s sex, defined as sex of the child and categorized as (i) male, (ii) female; (3) birth interval, defined as interval between two births, categorized as (i) less than 24 months, (ii) 24 months or longer; (4) maternal education, categorized as (i) no education, (ii) primary, (iii) secondary and higher; (5) maternal age, grouped as (i) 15–24 years, (ii) 25–34 years, and (iii) 35 years and older; (6) religious affiliation, categorized as (i) Christianity (ii) Islam (iii) and other; (7) place of delivery, defined as place where child was delivered, and categorized as (i) home, (ii) health facility; and (8) wealth index, categorized as (i) poorest, (ii) poorer, (iii) middle, (iv) richer, (v) richest. The wealth index is the proxy indicator for household socioeconomic status. This was derived from the scores allocated to various household possessions. Wealth index was applied in this analysis as a composite index and an indicator of the socioeconomic status of households. This was because the Demographic and Health Survey generally does not collect information on household income or wealth.
Infant and child mortality in Nigeria
171
Statistical analysis Three levels of analysis – univariate, bivariate and multivariate – were employed in data analysis. At the univariate level, descriptive analysis was done and sample characteristics were presented in percentages to show the distribution of respondents by the selected variables. At the bivariate level, cross-tabulation was done and Pearson’s chi-squared test was performed to examine relationship between the outcome variables and the selected independent variables. At the multivariate level of analysis, a two-level multilevel Cox proportional regression analysis was performed to examine the effects of individual- and community-level characteristics on child survival during infancy and childhood, and to determine the extent to which characteristics at the individual and community levels explain regional variations in infant and child mortality in Nigeria. The multilevel Cox proportional hazards model (survival analysis) was employed for multivariate analysis. This was done for two reasons. First, Cox proportional hazards regression analysis is appropriate for analysis of survival data. It is particularly appropriate for handling censored observations. In social science research, censoring occurs when the value of an observation is not fully known. Cox regression analysis allows for the inclusion of censored data and it models censored time-until-event data as a dependent variable where it can be assumed that the covariates have a multiplying effect on hazard rates. Using Cox proportional hazards regression analysis, both the occurrence of childhood mortality and the time when the child died were combined to generate the outcome variables. The second reason for using multilevel Cox proportional hazards model was to account for the hierarchical structure of the data. The assumption is that children and their mothers (individuals) are nested within households, while households are in turn nested within communities (Harttgen & Misselhorn, 2006). This suggests that children in households with similar characteristics can have different health outcomes when residing in different communities with different characteristics. Using the multilevel Cox proportional hazards model, the probability of childhood death was regarded as the hazard. The hazard was modelled using the following equations: HðtÞ ln ð1Þ ¼ b1 X 1 þ b 2 X 2 þ b 3 X3 þ þ b k Xk ; H0 ðtÞ where X1 . . . Xk are a collection of explanatory variables and H0(t) is the baseline hazard at time t, representing the hazard for a person with the value 0 for all the explanatory variables. By dividing both sides of equation (1) by H0(t) and taking logarithms, equation (1) becomes: HðtÞ ¼ b1 X1 þ b 2 X2 þ b3 X3 þ þ b k Xk ; ln ð2Þ H0 ðtÞ where H(t)/H0(t) is regarded as the hazard ratio. The coefficients bi . . . bk are estimated by Cox regression. To estimate both the fixed and random effects in the multilevel survival analysis, it could be assumed that the hazards of any two units are proportional (Rabe-Hesketh et al., 2004) and this can be modelled as: hij ðtÞ ¼ h0 ðtÞ expðvij Þ:
ð3Þ
172
S. A. Adedini et al.
In equation (3) above, there are two levels (the two subscripts): i represents the level 1 units (individuals), j stands for the level 2 units (communities) and nij denotes the linear predictor of the generalized linear latent and mixed model (GLLAMM). Further, to examine how individual- and community-levels determinants influence survival chances during infancy and childhood, separate models were fitted for infant mortality and child mortality. Fourteen models were fitted in all (seven models each for the two outcome variables). The first model (Model 0 or empty model) contained no explanatory variables, but was fitted to decompose the total variance into its individual- and community-level components. The second model (Model 1) considered only the region of residence covariate in order to examine the independent influence of the region where children were born or raised on their survival chance. The third model (Model 2) incorporated the child-level variables into the multilevel analysis. While the fourth model (Model 3) incorporated the mother-level variables, the fifth model (Model 4) considered only the community-level variables in order to examine the effect of community-level factors on child survival, independent of other factors. The sixth model (Model 5) is the full model that incorporated all the selected variables into the multilevel analysis. The seventh model (Model 6) is the final model. Fitting the final model involved two steps. First, stepwise survival analysis was done to determine the key variables associated with infant and child mortality. Second, all the variables selected from the stepwise Cox regression models were incorporated into the multilevel modelling. All analysis was done using Stata (version 11.2). GLLAMM – a downloadable program and implementable in Stata – was used to conduct all the multilevel analyses. Fixed effects and random effects, which are important concepts in multilevel analysis, were employed in results interpretation. While fixed effects are used to model associations, random effects are useful in modelling variations (Merlo et al., 2005, 2006). Conventionally, measures such as regression coefficients, odds ratios and hazard ratios are useful measures of association, but these give no information on the health variations within and between populations. Thus in multilevel modelling, measures of variations such as variance partition coefficient (or intra-class correlation) and proportional change in variance are good measures that provide good understanding of contextual determinants of individual health (Merlo et al., 2005). In this study, measures of variation represent the extent to which children raised in the same neighbourhood or community are exposed to the same situations such as availability (or non-availability) of health services, medical personnel, electricity, drinkable water and others. Measures of association (i.e. fixed effects) were expressed in this study as hazard ratios (HRs) and p-values. The random effects, which measure variations in infant and child mortality across communities, were expressed in this study as intra-class correlation (ICC) (or variance partition coefficient, VPC), and proportional change in variance (PCV). The intra-class correlation is an important measure of the relatedness of clustered data within community or household units (Antai, 2011b). The VPC was calculated in this study using the linear threshold model method whereby VPC corresponds to the intra-class correlation (Merlo et al., 2005). Hence, the VPC was computed using: ¼ ð 2 = 2 þ ð 2 =3ÞÞ;
Infant and child mortality in Nigeria 2
173 2
where, r is the ICC, d is the variance at the community level, p /3 ¼ 3.29 and represents the fixed variance at individual level (Merlo et al., 2005). The precision of random effects was determined by the standard error (SE) of the covariates. To determine the goodness-of-fit of the consecutive models, regression diagnostic was done using Akaike Information Criteria (AIC). Boco (2010) noted that a lower value of AIC indicates a better fit. Results Individual-level characteristics by region of residence The distribution of the study sample by individual-level characteristics and according to region of residence is presented in Table 1. Huge differences exist in the selected characteristics between regions. With the exception of child’s sex, all the selected characteristics vary significantly across the regions of residence (p < 0.001). With respect to
Table 1. Percentage distribution of child- and mother-level characteristics by region of residence, Nigerian DHS 2008 Variable/category Region of residence Child’s sex Male Female Birth order First birth 2–4 5þ Birth interval
of Biosocial Science:
Email alerts: Click here Subscriptions: Click here Commercial reprints: Click here Terms of use : Click here
REGIONAL VARIATIONS IN INFANT AND CHILD MORTALITY IN NIGERIA: A MULTILEVEL ANALYSIS SUNDAY A. ADEDINI, CLIFFORD ODIMEGWU, EUNICE N. S. IMASIKU, DOROTHY N. ONONOKPONO and LATIFAT IBISOMI Journal of Biosocial Science / Volume 47 / Issue 02 / March 2015, pp 165 - 187 DOI: 10.1017/S0021932013000734, Published online: 10 January 2014
Link to this article: http://journals.cambridge.org/abstract_S0021932013000734 How to cite this article: SUNDAY A. ADEDINI, CLIFFORD ODIMEGWU, EUNICE N. S. IMASIKU, DOROTHY N. ONONOKPONO and LATIFAT IBISOMI (2015). REGIONAL VARIATIONS IN INFANT AND CHILD MORTALITY IN NIGERIA: A MULTILEVEL ANALYSIS. Journal of Biosocial Science, 47, pp 165-187 doi:10.1017/S0021932013000734 Request Permissions : Click here
Downloaded from http://journals.cambridge.org/JBS, IP address: 131.187.94.93 on 18 Mar 2015
J. Biosoc. Sci., (2015) 47, 165–187, 6 Cambridge University Press, 2014 doi:10.1017/S0021932013000734 First published online 10 Jan 2014
REGIONAL VARIATIONS IN INFANT AND C H I L D M O R T A L I T Y I N NI G E R I A : A M U L T IL E V E L A N A L Y S I S SUNDAY A. ADEDINI*†1, CLIFFORD ODIMEGWU*, EUNICE N. S. IMASIKU*‡, DOROTHY N. ONONOKPONO*§ and LATIFAT IBISOMI* *Demography and Population Studies Programme, Schools of Public Health and Social Sciences, University of the Witwatersrand, Johannesburg, South Africa, †Demography and Social Statistics Department, Obafemi Awolowo University, Ile-Ife, Nigeria, ‡Department of Geography, University of Zambia, Lusaka, Zambia and §Department of Sociology and Anthropology, University of Uyo, Nigeria Summary. There are substantial regional disparities in under-five mortality in Nigeria, and evidence suggests that both individual- and community-level characteristics have an influence on health outcomes. Using 2008 Nigeria Demographic and Health Survey data, this study (1) examines the effects of individual- and community-level characteristics on infant/child mortality in Nigeria and (2) determines the extent to which characteristics at these levels influence regional variations in infant/child mortality in the country. Multilevel Cox proportional hazard analysis was performed on a nationally representative sample of 28,647 children nested within 18,028 mothers of reproductive age, who were also nested within 886 communities. The results indicate that community-level variables (such as region, place of residence, community infrastructure, community hospital delivery and community poverty level) and individual-level factors (including child’s sex, birth order, birth interval, maternal education, maternal age and wealth index) are important determinants of infant/child mortality in Nigeria. For instance, the results show a lower risk of death in infancy for children of mothers residing in communities with a high proportion of hospital delivery (HR: 0.70, p < 0.05) and for children whose mothers had secondary or higher education (HR: 0.84, p < 0.05). Although community factors appear to influence the association between individual-level factors and death during infancy and childhood, the findings consistently indicate that community-level characteristics are more important in explaining regional variations in child mortality, while individual-level factors are more important for regional variations in infant mortality. The results of this study underscore the need to look beyond the influence of individual-level factors in addressing regional variations in infant and child mortality in Nigeria. 1
Corresponding author. Email: [email protected]
165
166
S. A. Adedini et al. Introduction
Despite modest improvements in child health outcomes during the 20th century, infant and child mortality rates remain unacceptably high in the sub-Saharan African countries. Globally, around 7 million under-five deaths were recorded in 2011 (UNICEF, 2012). Sub-Saharan Africa is a major contributor to this statistic as more than two in five under-five deaths occur in the region (Black et al., 2003; Rutherford et al., 2010). Scholars have attempted to examine the factors influencing this. For instance, studies have established a significant relationship between infant and child mortality and individual-level characteristics such as maternal education, wealth status and other socioeconomic characteristics (Adetunji, 1995; Zaba & David, 1996; Lawoyin, 2001; Buor, 2002; Odimegwu, 2002; Fayeun & Omololu, 2011). Other studies have shown that attributes of the community context tend to influence the health outcomes of individuals (Sastry, 1996; Uthman, 2008; Babalola & Fatusi, 2009; Adekanmbi et al., 2013). This suggests that living in an economically and socially deprived community or neighbourhood could bring about an increase in mortality risk. Nigeria – the most populous country in Africa – is characterized by socially and economically advantaged and disadvantaged regions. The country is geographically, religiously, socially, ecologically and economically diverse. This has led to varied disease exposures and different health outcomes (Wall, 1998; Lawoyin, 2001; Grais et al., 2007). From the arid northern region of the country to the savannah west, from the predominantly Islamic north-west to the south-east vastly dominated by Christians, the country is highly heterogeneous and diverse. As a result, there is a huge diversity in the regional environment, cultural practices (Antai et al., 2009), health-seeking practices (Babalola & Fatusi, 2009), socioeconomic status (Aremu et al., 2011) and the political milieu. To date, Nigeria has a very high rate of under-five mortality, as about one in every six children born in the country dies before the age of five (NPC & ICF Macro, 2009). Worse still, infant and child mortality rates vary substantially from one region of the country to the other. For instance, under-five mortality rate ranges from 89 per 1000 live births in the south-west to 222 per 1000 live births in the north-east. Besides, Nigeria is not making sufficient progress towards the attainment of Millennium Development Goal four (MDG-4), as the under-five mortality rate in the country for the 1998–2003 period was 201 per 1000 live births, while the rate marginally declined to 157 per 1000 live births during the 2004–2008 period (NPC & ORC Macro, 2004; NPC & ICF Macro, 2009). In addition, as noted earlier, Nigeria is by far the most populous country in Africa and has a very huge childhood population. According to the 2006 population and housing census, Nigerian’s population is 140,431,790 and the population of the under-five children is 16.1% of the total population (NPC & ICF Macro, 2009). Furthermore, despite the improvement in medical technology, reports by NPC & ICF Macro (2009) indicate that Nigeria is still faced with several health challenges. These include high rates of under-five mortality (157 per 1000 live births), high teenage pregnancy (23% of young women aged 15–19 have already given birth), many poor pregnancy outcomes (such as stillbirth, spontaneous abortion and low birth weight), poor survival chances for the newborn and high unmet need for family planning (20% of currently married women have an unmet need for contraception in the country).
Infant and child mortality in Nigeria
167
Also, the percentage of births assisted by unskilled birth attendants is 60% in the country. The list of the country’s public health problems is endless. While many studies have been conducted on childhood mortality in Nigeria, as in other developing countries, Nigerian studies have paid more attention to the influence of individual-level attributes, and less attention to community-level determinants of childhood mortality. However, the literature shows that knowledge about the determinants of childhood mortality at the individual level is insufficient to address the problem (Sastry, 1996; Whitworth & Stephenson, 2002; Griffiths et al., 2004; Harttgen & Misselhorn, 2006; Antai, 2011b). This is because the contextual characteristics of the community or neighbourhood where children are born or raised tend to modify childlevel and mother-level characteristics and therefore affect children survival chances. In addition, past Nigerian studies that have examined community determinants of child survival have focused mainly on ages 0–59 months, thereby ignoring what the effects of community characteristics on child survival are during infancy (i.e. age 0–11 months) and childhood (12–59 months). To this end, this study aims to (1) examine the effects of individual- and community-level characteristics on child survival during infancy and childhood, and (2) determine the extent to which characteristics at the individual and community levels explain regional variations in infant and child mortality in Nigeria. Theoretical background From a theoretical standpoint, Mosley and Chen’s 1984 model on the proximate causes of childhood mortality establishes a relationship between child survival and determinants at various levels of operation: individual, household and community levels (WHO, 2003). Diez-Roux et al. (2001) posited that the physical and social characteristics of the neighbourhood where a person lives may affect health and health-related behaviour. Galster’s (2010) work on the mechanisms of neighbourhood effects theory observed a link between residential environment and the health outcomes of individual adults and children residing in such environment or community. The effect of community context, where children are born or raised, on their survival chances has been widely recognized (Sastry, 1997b; Omariba et al., 2007; Antai, 2009). Evidence suggests that living in an economically and socially deprived community or neighbourhood is associated with an increased risk of under-five mortality (Aremu et al., 2011). For instance, children born or raised in a community that lacks electricity, good drinking water and health facilities are likely to suffer from the same deprivation that can directly or indirectly influence their health outcomes. Further, Manda (2001) opined that Demographic and Health Surveys (DHSs) often collect birth history data that are clustered at the household and community levels. Sastry (1997b) also maintained that most demographic surveys conducted in developing countries often collect survival data that are clustered at both family and community levels. Omariba and colleagues (2007) argued that since DHSs collect child survival data from individual mothers in sampled households, then the children of those mothers cannot be regarded as independent observations. This is as a result of natural clustering or as a result of the data collection procedures (i.e. two-stage cluster sampling design) used in DHS data collection (Sastry, 1997b). The children from sampled households who are also nested within the same individual mother’s level tend to share similar characteristics and common genetic
168
S. A. Adedini et al.
factors. This is also true of children from the same community. Individuals in the same community are likely to be more homogenous than those from different communities. Similarities are expected in the health outcomes of children who are exposed to the same environmental conditions. By contrast, differences are expected in health outcomes of children raised in different communities due to differences in community characteristics (Harttgen & Misselhorn, 2006). A distinction is drawn between children living in a relatively better-off neighbourhood and those living in a relatively worse-off neighbourhood (Macintyre et al., 2002). Children living in two different households with similar socioeconomic characteristics can suffer different mortality risks if they are from two contrasting communities. Sastry (1996) contended that community characteristics can exacerbate or mitigate mortality risks of individuals depending on the environmental context the individuals find themselves. Griffiths (2004) also argued that community services and levels of infrastructural development of a community are capable of amplifying or reducing mortality risks among the children. This is because an individual child resident in a household unit, which in turn is located within a community, is exposed to various levels (within the societal hierarchy) that either directly or indirectly influence his or her survival chances. In addition, Whitworth & Stephenson (2002) maintained that two neonates with similar characteristics may suffer different neonatal mortality risks because of the community contextual effects. The authors argued that these differentials in mortality risks may be as a result of differences in the provision of antenatal and obstetric health care or the effects of environmental conditions the children are exposed to. Also, individuals residing in the same community tend to share similar preferences, cultural practices, values and customs. The reason is that individuals with similar tastes and values tend to cluster and live together (Sastry, 1997a). All this clustering and living together of people with common norms, values, identities and cultural practices, as well as spatial inequality in infrastructural development (Antai, 2011b), has a direct or indirect effect on the health outcomes of under-five children and this often brings about differentials in health outcomes between communities, particularly communities with contrasting characteristics. Motivated by the emerging interest in the study of effects of community or neighbourhood context on health outcomes in developing countries, this paper seeks to examine the extent to which characteristics at individual and community levels influence regional variations in infant and child mortality in Nigeria. Because previous attempts at examining regional disparities in child survival in Nigeria have focused mainly on the period within age 0–59 months, the main focus of the present analysis was the regional disparities in child survival during the two distinct periods of infancy (0–11 months) and childhood (12–59 months). Data and Methods Data source This study utilizes the birth recode of the 2008 Nigeria Demographic and Health Survey (NDHS) data. The survey elicited information on demographic and health indicators. The primary sampling unit (PSU), which was regarded as a cluster for the
Infant and child mortality in Nigeria
169
2008 NDHS, was defined on the basis of enumeration areas (EAs). The sample for the survey was selected using a stratified two-stage cluster design consisting of 888 clusters (NPC & ICF Macro, 2009). In all, a nationally representative sample of 36,800 households was selected for the survey. The community-level variables were measured at the level of the PSU, which serves as a proxy for community or neighbourhood. The PSUs refer to small and administratively defined areas and one PSU is a cluster consisting of at least 80 fairly homogenous households or units. In the 2008 NDHS, data were collected in 886 PSUs, while two PSUs could not be accessed due to disturbances in the areas. A full report of the data collection procedures for the 2008 NDHS is available elsewhere (NPC & ICF Macro, 2009). In the 2008 NDHS, birth history data were collected from 33,385 women aged 15–49 years. These included sex of child, month and year of child’s birth, child’s survivorship status, child’s current age and age at death if the child had died. Analysis for the present paper was restricted to 28,647 live-born children produced by 18,028 women within the five years before the survey. Analysis was restricted to births within the five years preceding the survey to obtain a picture of the current situations in the various geo-political regions of the country. However, severe omission of births and deaths as well as displacement of dates of those events could seriously affect mortality estimates. Meanwhile, the data quality assessment of the 2008 NDHS indicates that the percentage of missing information regarding births, deaths, birth dates as well as age at death across various characteristics of mothers such as mother’s education and region of residence only varied between less than 1% and 3% (NPC & ICF Macro, 2009). This suggests that there was neither a serious under-reporting of infant and child deaths nor serious displacement of information on such vital events that could seriously affect the mortality estimates across regions or across any other mother-level characteristic. In addition, to ensure national representativeness, weighting factors were applied to adjust for oversampling of some locations and under-sampling of others in the Demographic and Health Survey. Ethical considerations This study was based on secondary analysis of an existing dataset with all participant identifiers removed. The survey instruments received ethical approval from the National Ethics Committee in the Federal Ministry of Health, Abuja, Nigeria, and from the Ethics Committee of the Opinion Research Corporation of Macro International Inc., Calverton, MD, USA. Permission to use the 2008 Nigeria DHS data for this study was obtained from ICF Macro Inc. Outcome variables The outcome variables for this study were the risks of death in infancy or childhood, measured as the duration of survival since birth in months. These are defined as either the risk of a child dying between birth and first birthday (infant mortality) or between age 12 and 59 months (child mortality). Analysis was child-based and restricted to the live births in the 5 years before the survey. Hence, all children born within the 5 years before the survey date were included in the analysis. The children’s survival status and the age
170
S. A. Adedini et al.
at death in months (if the child had died) or the last month they were known to be alive (if the child was still living at the time of the survey) were combined to generate the outcome variables for the survival analysis. Taking the two durations (0–11 months and 12–59 months) into consideration, children known to have died (i.e. non-censored) were regarded as the cases, whereas children who were still alive at the time of the survey were treated as right-censored (as appropriate for infant and child mortality). Exposure variables Community contextual factors. As established in the reviewed literature, the community contextual characteristics of interest in this study included: (1) place of residence, categorized as (i) urban, (ii) rural; (2) region of residence, defined as the regions where the children were raised, and categorized as (i) South-West, (ii) North-Central, (iii) North-East, (iv) North-West, (v) South-East, and (vi) South-South; (3) community poverty level, defined as the average poverty level in the community, and categorized as (i) low, (ii) middle and (iii) high. Other contextual characteristics were (4) community maternal level of education, defined as the proportion of mothers who had at least secondary level of education in the community and categorized as (i) low, (ii) middle, (iii) high; (5) community hospital delivery, defined as proportion of mothers who had hospital delivery in the community, and categorized as (i) low, (ii) middle (iii) high; (6) community prenatal care by skilled provider, defined as the proportion of mothers who attended prenatal care by a skilled health provider in the community and categorized as (i) low, (ii) middle (iii) high; (7) proportion with electric connection in the community, defined as the proportion of mothers from households with electric connection in the community and categorized as (i) low, (ii) middle (iii) high; and (8) proportion with piped water in the community, defined as the proportion of mothers from households that had piped water as source of drinking water in the community and categorized as (i) low, (ii) middle (iii) high. Individual-level characteristics. Based on the reviewed literature, the important individual-level (i.e. child-level and mother-level) characteristics considered in this study were as follows: (1) birth order, defined as birth order of the child and categorized as (i) first births, (ii) 2–4 birth order, (iii) birth order 5þ; (2) child’s sex, defined as sex of the child and categorized as (i) male, (ii) female; (3) birth interval, defined as interval between two births, categorized as (i) less than 24 months, (ii) 24 months or longer; (4) maternal education, categorized as (i) no education, (ii) primary, (iii) secondary and higher; (5) maternal age, grouped as (i) 15–24 years, (ii) 25–34 years, and (iii) 35 years and older; (6) religious affiliation, categorized as (i) Christianity (ii) Islam (iii) and other; (7) place of delivery, defined as place where child was delivered, and categorized as (i) home, (ii) health facility; and (8) wealth index, categorized as (i) poorest, (ii) poorer, (iii) middle, (iv) richer, (v) richest. The wealth index is the proxy indicator for household socioeconomic status. This was derived from the scores allocated to various household possessions. Wealth index was applied in this analysis as a composite index and an indicator of the socioeconomic status of households. This was because the Demographic and Health Survey generally does not collect information on household income or wealth.
Infant and child mortality in Nigeria
171
Statistical analysis Three levels of analysis – univariate, bivariate and multivariate – were employed in data analysis. At the univariate level, descriptive analysis was done and sample characteristics were presented in percentages to show the distribution of respondents by the selected variables. At the bivariate level, cross-tabulation was done and Pearson’s chi-squared test was performed to examine relationship between the outcome variables and the selected independent variables. At the multivariate level of analysis, a two-level multilevel Cox proportional regression analysis was performed to examine the effects of individual- and community-level characteristics on child survival during infancy and childhood, and to determine the extent to which characteristics at the individual and community levels explain regional variations in infant and child mortality in Nigeria. The multilevel Cox proportional hazards model (survival analysis) was employed for multivariate analysis. This was done for two reasons. First, Cox proportional hazards regression analysis is appropriate for analysis of survival data. It is particularly appropriate for handling censored observations. In social science research, censoring occurs when the value of an observation is not fully known. Cox regression analysis allows for the inclusion of censored data and it models censored time-until-event data as a dependent variable where it can be assumed that the covariates have a multiplying effect on hazard rates. Using Cox proportional hazards regression analysis, both the occurrence of childhood mortality and the time when the child died were combined to generate the outcome variables. The second reason for using multilevel Cox proportional hazards model was to account for the hierarchical structure of the data. The assumption is that children and their mothers (individuals) are nested within households, while households are in turn nested within communities (Harttgen & Misselhorn, 2006). This suggests that children in households with similar characteristics can have different health outcomes when residing in different communities with different characteristics. Using the multilevel Cox proportional hazards model, the probability of childhood death was regarded as the hazard. The hazard was modelled using the following equations: HðtÞ ln ð1Þ ¼ b1 X 1 þ b 2 X 2 þ b 3 X3 þ þ b k Xk ; H0 ðtÞ where X1 . . . Xk are a collection of explanatory variables and H0(t) is the baseline hazard at time t, representing the hazard for a person with the value 0 for all the explanatory variables. By dividing both sides of equation (1) by H0(t) and taking logarithms, equation (1) becomes: HðtÞ ¼ b1 X1 þ b 2 X2 þ b3 X3 þ þ b k Xk ; ln ð2Þ H0 ðtÞ where H(t)/H0(t) is regarded as the hazard ratio. The coefficients bi . . . bk are estimated by Cox regression. To estimate both the fixed and random effects in the multilevel survival analysis, it could be assumed that the hazards of any two units are proportional (Rabe-Hesketh et al., 2004) and this can be modelled as: hij ðtÞ ¼ h0 ðtÞ expðvij Þ:
ð3Þ
172
S. A. Adedini et al.
In equation (3) above, there are two levels (the two subscripts): i represents the level 1 units (individuals), j stands for the level 2 units (communities) and nij denotes the linear predictor of the generalized linear latent and mixed model (GLLAMM). Further, to examine how individual- and community-levels determinants influence survival chances during infancy and childhood, separate models were fitted for infant mortality and child mortality. Fourteen models were fitted in all (seven models each for the two outcome variables). The first model (Model 0 or empty model) contained no explanatory variables, but was fitted to decompose the total variance into its individual- and community-level components. The second model (Model 1) considered only the region of residence covariate in order to examine the independent influence of the region where children were born or raised on their survival chance. The third model (Model 2) incorporated the child-level variables into the multilevel analysis. While the fourth model (Model 3) incorporated the mother-level variables, the fifth model (Model 4) considered only the community-level variables in order to examine the effect of community-level factors on child survival, independent of other factors. The sixth model (Model 5) is the full model that incorporated all the selected variables into the multilevel analysis. The seventh model (Model 6) is the final model. Fitting the final model involved two steps. First, stepwise survival analysis was done to determine the key variables associated with infant and child mortality. Second, all the variables selected from the stepwise Cox regression models were incorporated into the multilevel modelling. All analysis was done using Stata (version 11.2). GLLAMM – a downloadable program and implementable in Stata – was used to conduct all the multilevel analyses. Fixed effects and random effects, which are important concepts in multilevel analysis, were employed in results interpretation. While fixed effects are used to model associations, random effects are useful in modelling variations (Merlo et al., 2005, 2006). Conventionally, measures such as regression coefficients, odds ratios and hazard ratios are useful measures of association, but these give no information on the health variations within and between populations. Thus in multilevel modelling, measures of variations such as variance partition coefficient (or intra-class correlation) and proportional change in variance are good measures that provide good understanding of contextual determinants of individual health (Merlo et al., 2005). In this study, measures of variation represent the extent to which children raised in the same neighbourhood or community are exposed to the same situations such as availability (or non-availability) of health services, medical personnel, electricity, drinkable water and others. Measures of association (i.e. fixed effects) were expressed in this study as hazard ratios (HRs) and p-values. The random effects, which measure variations in infant and child mortality across communities, were expressed in this study as intra-class correlation (ICC) (or variance partition coefficient, VPC), and proportional change in variance (PCV). The intra-class correlation is an important measure of the relatedness of clustered data within community or household units (Antai, 2011b). The VPC was calculated in this study using the linear threshold model method whereby VPC corresponds to the intra-class correlation (Merlo et al., 2005). Hence, the VPC was computed using: ¼ ð 2 = 2 þ ð 2 =3ÞÞ;
Infant and child mortality in Nigeria 2
173 2
where, r is the ICC, d is the variance at the community level, p /3 ¼ 3.29 and represents the fixed variance at individual level (Merlo et al., 2005). The precision of random effects was determined by the standard error (SE) of the covariates. To determine the goodness-of-fit of the consecutive models, regression diagnostic was done using Akaike Information Criteria (AIC). Boco (2010) noted that a lower value of AIC indicates a better fit. Results Individual-level characteristics by region of residence The distribution of the study sample by individual-level characteristics and according to region of residence is presented in Table 1. Huge differences exist in the selected characteristics between regions. With the exception of child’s sex, all the selected characteristics vary significantly across the regions of residence (p < 0.001). With respect to
Table 1. Percentage distribution of child- and mother-level characteristics by region of residence, Nigerian DHS 2008 Variable/category Region of residence Child’s sex Male Female Birth order First birth 2–4 5þ Birth interval
